The article outlines the conceptual foundations of new trends in control theorythat have been intensively developing in recent years. Unlike classical controltheory, which was formed in the last century and is based on well-known mathematical models of controlled processes in the form of local equations, new approaches to linear stationary systems use input-output relations that follow directly from the Cauchy formula for both continuous and discrete systems. On thebasis of the same description, it is possible to substantiate and obtain the socalled data-based models, which are directly linked to data that form, at the observation intervals, the trajectories of already implemented past processes andfuture ones, for which control is to be synthesized. This approach is focusedprimarily on finding control from the prediction model. At the same time, thecurrent measurements carried out at the plant make it possible to implementfeedback and, in case of discrepancies between the forecast and the real process,to correct the predictive control, i.e. such a way to stabilize it. Control by trajectory prediction model allows to exclude model identification by trajectory data,and control directly on their base. Since the data contain errors, the most important issue in the considered approach is the robustness of the chosen control.A large number of published works are dedicated to this problem, where theguaranteed approach, focused on the worst-case in the data, is the most in demand. In most cases, control synthesis is reduced to solving various optimizationproblems, mainly on the finite prediction horizon. Considerable attention in thearticle is paid to methods for solving synthesis problems based on SVD decomposition. To reduce the complexity of the tasks to be solved, it is proposed to reduce it to terminal control on the horizon of a short duration. Then an iterativecontrol strategy is implemented, which, due to feedback, ensures the feasibilityof the global control goal.
Read full abstract